These are chat archives for atomix/atomix

27th
Mar 2018
Jordan Halterman
@kuujo
Mar 27 2018 05:53
I’ll be back on Atomix after I give a talk in the morning :-)
Johno Crawford
@johnou
Mar 27 2018 06:19
Recording it?
Jordan Halterman
@kuujo
Mar 27 2018 06:30
should be, but it’s not on Atomix
not directly at least
Jordan Halterman
@kuujo
Mar 27 2018 07:53
It’s on “in-service” partitioned rolling upgrades of ONOS controllers
Although I will be doing a talk on Atomix 2.1 soon
Probably in a few weeks
rbondar
@rbondar
Mar 27 2018 13:57

guys , atomix 1 supported groups.

atomix.getGroup("my-group").thenAccept(group -> {
  ...
});

What is the replacement in atomix 2?

rbondar
@rbondar
Mar 27 2018 14:28
is it ok to use the same directory for several different atomix clusters in one JVM ?
Jordan Halterman
@kuujo
Mar 27 2018 17:10
no it’s not… those directories store all the Raft log files, and you’ll end up having multiple. nodes reading/writing the same files and see some really crazy behavior
rbondar
@rbondar
Mar 27 2018 17:11
got it, separated directories for different clusters
Jordan Halterman
@kuujo
Mar 27 2018 17:11
the replacement for groups is a variety of new services and primitives that provide the same semantics with better algorithms: LeaderElector, LeaderElection, ClusterService
they use phi accrual failure detectors and are much more efficient
rbondar
@rbondar
Mar 27 2018 17:14
thank you
guys, when we can start test CONSENSUS nodes? i'm really worry of current behavior of CLIENT and DATA nodes - when CLIENT node is stopped DATA nodes still try to connect to the stopped CLIENT node.
time delay failure threshold does not trigger any action.
Jordan Halterman
@kuujo
Mar 27 2018 18:06
DATA nodes try to connect to CLIENT nodes, not vice versa?
I thought it was the other way around
Jordan Halterman
@kuujo
Mar 27 2018 19:00
@rbondar

Also, DATA nodes try to connect to the CLIENT nodes forever?

They should attempt to connect and detect a failure then stop connecting. Just double checked. Can you provide a reproducer? Logs?

Jordan Halterman
@kuujo
Mar 27 2018 20:51
I’m unable to reproduce it
Jordan Halterman
@kuujo
Mar 27 2018 21:21
In unit tests or in containers @rbondar
Also, I suppose we will merge the CORE node changes after it’s tested in containers